Research On Intelligent Detection Technology Of Underground Drainage Pipeline Defects Based On Deep Learning

With the development of moderm cities,the scale of cities is expanding ancd more and more residents are living.T,he drainage pipelines buried deep in the urban underground in the early stage are already overburdened,which attracts more and more attention.At present,in the field of Engineering application,the defect of drainage pipeline is mainly identified by m anual naked eye,which is time-consuming and laborious,and has large subjective error.Therefore,it is of great practical significance to carry out the research of intelligent defect identification of drainage pipeline.Pipeline defect has many kinds and no obvious difference,which makes it very difficult to classify and recognize the image and segment the accurate image.At present,the intelligent detection and recognition technology of pipeline defect is still in its infancy.In order to help engineers quickly detect the defect of drainage pipeline,analyze the appearance characteristics(size and severity of defect,etc.),and combine with the actual needs of the project(defect category,defect location,evaluation of the amount of defect repair works,etc.),an intelligent inspection of the defect of underground drainage pipeline based on in-depth learning is proposed.Through the improved AlexNet network,the automatic classification of ten types of defects,such as drainage pipe falling off,cracking,cracking,precipitation,scum,scaling and corrosion,tree root,dislocation,obstacle and branch pipe concealed connection,is realized.The precise segmentation and labeling of the specific location of defects is completed by using SegNet network,and the main conclusions are as follows.1.By analyzing the current situation of drainage pipelines in China,ten typical defects of drainage pipelines(shedding,cracking,cracking,precipitation,scum,scaling and corrosion,tree roots,dislocation of dislocations,obstacles,branch pipe hidden connection)are obtained according to the probability of occurrence of defects,hazards and defect characteristics,and the image features of each defect are introduced.Make an in-depth analysis.2.On the basis of the analysis of the characteristics of various defect images,the deep learning image processing technology is introduced to classify and collect defect images,preprocess and make classification labels.The high-quality data sets are generated,and the defect classification model based on AlexNet network is constructed,which realizes the intelligent recognition and classification of drainage pipeline defects.The factors affecting the accuracy of the deep learning network(learning rate,iteration times and noise parameters)are studied in depth,and the optimal combination of parameters is obtained.Finally,the generated data sets are input into the optimal combination of parameters of AlexNet network for training,and a stable network model of drainage pipeline defect recognition is obtained.The classification accuracy of the trained model is 92%.The experimental results show that the depth convolution neural network can obtain better image information,extract more effective image features,and the recognition accuracy can be applied to engineering practice.3.On the basis of intelligent recognition and classification of various defects,in order to further clarify the specific location,severity,and estimate the quantity and cost of defect treatment,a data set was established by using phofoshop software to label each defect location in defect images,and an automatic defect area labeling model based on SegNet network was constructed.The defect area of drainage pipeline can be marked automatically.The trained model was used to label the defect area of drainage pipeline,and the accuracy rate was 800-%.The experimental results show that the depth convolution neural network can accurately distinguish and mark the specific location of different defects,and the pixel accuracy can also meet the general engineering needs.4.Based on the successful construction of drainage pipeline defect recognition classification network and intelligent labeling network,combined with the specific drainage pipeline network endoscopy detection and pipeline network evaluation project of a survey and design institute,the network was applied in practice.The classification and identification accuracy of drainage pipeline defect reached 91%,and the regional segmentation and labeling accuracy reached 80%.From the results,it can be concluded that the application of deep learning technology in engineering projects is feasible.